This research developed and validated QSPKR models for predicting in-vivo human, systemic biologically relevant PK properties (i.e., reflecting the disposition of the unbound drug) of four, preselected, pharmacological classes of drugs, namely, benzodiazepines (BZD), neuromuscular blocking agents (NMB), triptans (TRP) and class III antiarrhythmic agents (AAR), as well as PK allometric scaling (PK-AS) models for BZD and NMB, using pertinent human and animal systemic PK information (fu, CLtot, Vdss and fe) from published literature. Overall, lipophilicity (logD7.4) and molecular weight (MW) were found to be the most important and statistically significant molecular properties, affecting biologically relevant systemic PK properties, and the observed relationships were mechanistically plausible: For relatively small MW and lipophilic molecules, (e.g., BZD), an increase in logD7.4 was associated with a decrease in fu, an increase in Vdssu and CLnonrenu, suggesting the prevalence of nonspecific hydrophobic interactions with biological membranes/plasma proteins as well as hepatic partitioning/DME binding. Similar trends were observed in fu and Vdssu for intermediate to large MW, hydrophilic molecules (e.g., NMB). However, although similar trends were observed in fu and Vdssu for relatively hydrophilic, intermediate MW molecules (e.g., TRP), and a heterogeneous class (e.g., Class III AAR), logD7.4 and MW were found to be highly correlated, i.e., the indepdendent effects of logD7,4 and MW cannot be assessed NMB, TRP and Class III AAR show mechanistically diverse clearance pathways, e.g., hepatobiliary, extrahepatic, enzymatic/chemical degradation and renal excretion; therefore, effects of the logD7.4 and/or MW are note generalizable for any of the clearances across classes. PK-AS analyses showed that Vdssu and Vdss scaled well with body weight across animal species (including humans) for BZD. Overall, within the limitations of the methods (and the sample size), ‘acceptable’ predictions (i.e., within 0.5- to 2.0-fold error range) were obtained for Vdssu and Vdss for BZD (and fu correction resulted in improvement of the prediction); however, none of the CLtot predictions were acceptable, suggesting major, qualitative interspecies differences in drug metabolism, even after correcting for body weight (BW). NMB undergo little extravascular distribution owing to their relatively large MW and charged nature, and, as a result, a high percentage of acceptable predictions was obtained for Vdss (based on BW). Similarly, the prediction of CLren (based on BW and glomerular filtration rate, GFR) was acceptable, suggesting that NMB are cleared by GFR across species, and there are no interspecies differences in their tubular handling. On the other hand, CLtot (and/or CLnonren) could not be acceptably predicted by PK-AS, suggesting major differences in their clearance mechanisms across animal species.

This research explored quantitative relationships (QSPKR) between different molecular descriptors and pertinent, systemic PK properties for 14 calcium channel blockers (CCB). Physicochemical properties (PC) such as molecular weight (MW), molar volume (MV), calculated logP (clogP), pKa, calculated logD7.4 (clogD), % ionized at pH 6.3 and pH 7.4, hydrogen bond donors (HBD), hydrogen bond acceptors (HBA), and number of rotatable bonds (nRot) were chosen as possible predictor variables for systemic PK properties for CCB, obtained from pertinent literature, assessing the PK of CCB after intravenous administration to healthy humans. All PC properties and molecular descriptors were computed using ACD-solubility/DB 12.01. Total body clearance (CLtot), steady-state volume of distribution (Vdss), total area under the plasma concentration-time profile (AUCoo), terminal half-life (t1/2), and fraction of drug excreted unchanged in urine (fe), if available, were obtained or derived from original references, exclusively from IV studies that administered CCB to healthy human volunteers. Several articles focused on drug interactions with grapefruit juice or the impact of renal/hepatic dysfunction, and in such cases, data from the healthy control group were used. Each study was evaluated for study design, PK sampling schedule, bioanalytical and PK analysis methods before inclusion into the final database. The assumption of linear systemic PK was verified by assessing AUCoo versus (IV) dose. Plasma protein binding information was collected from in-vitro experiments to obtain the fraction unbound in plasma (fu). Unbound volume of distribution at a steady state (Vdssu), unbound total (CLtotu), renal (CLrenu), and non-renal clearance (CLnonrenu) were estimated and compared with the relevant physiological references for Vdssu (plasma volume, blood volume, extracellular and intracellular spaces, total body water and body weight) and for the unbound clearances (liver blood flow, renal plasma flow, and glomerular filtration rate, GFR). Final PK property values were obtained by averaging across available studies. The distribution of both PC and PK properties were evaluated, and correlation matrices amongst PC properties were constructed to assess for collinearity. If two PC descriptors were found to be collinear, i.e. r, ≥ 0.8, only one of them was used in the final univariate analysis. Finally, univariate linear regression of all PK variables versus each molecular descriptor was performed; any relationship with p<0.05 and r2≥0.30 was considered to be statistically significant. The PC properties of the final 14 CCB were reasonably normally distributed with few exceptions. Overall, CCBs are small (MW range of 316-496 Da), basic and lipophilic (logD7.4 range of 1.5-5.1) molecules. On the other hand, for the PK properties, the distributions were found to be skewed with high standard deviations. Thus, all PK variables (except fu) were log-transformed. Although CCB are mostly highly plasma protein bound (fu range of 0.2-20%), they are characterized by extensive extravascular tissue distribution (Vdss range of 0.6-20.4 l/kg) and high, mainly metabolic, clearance (CLtot range of 3.7-131.7 ml/min/kg). Clevidipine is the only CCB undergoing extensive, extra-hepatic ester hydrolysis, responsible for the highest CLtot value. Urinary excretion for CCB is negligible. Amlodipine is a PK outlier due to its high Vdss (20.4 l/kg) and low CLtot (6.9 ml/min/kg, due to low hepatic extraction) with fu of 2%. Therefore, the final QSPKR analysis was performed including, as well as excluding amlodipine. Excluding amlodipine, the relationship between fu and logD7.4 was negative and significant (r2 of 0.4, n=12). The relationships between CLtotu, CLnonrenu and CLrenu and logD7.4 were found to be positive and significant (r2 between 0.6-0.7, n=3-12); none of the other PC variables affected any of the clearance terms. Although the relationship between Vdssu and logD7.4 was not significant (r2 of 0.25, n=12), it showed the expected positive slope. In fact, after removing bepridil (the remaining outlier in Vdssu), the relationship with logD7.4 became statistically significant (r2=0.46, n=11). The QSPKR obtained in this study for CCB, with logD7.4 being the main PC determinant for systemic PK properties, were similar to those previously reported for opioids, β-adrenergic receptor ligands and benzodiazepines. However, slope estimates for the relationships of CLnonrenu and CLtotu as a function of logD7.4 for CCB were higher compared to these previously studied compounds, which showed higher sensitivity, most likely as a result of their higher lipophilicity. Overall, lipophilicity measured as logD7.4 was found to be a statistically significant and plausible PC determinant for the biologically relevant systemic PK properties for CCB and other classes of drugs.

<p style="text-align: justify;">Reverzno-faznom tečnom hromatografijom pod visokim pritiskom primenom dva sistema<br />rastvarača ispitano je pona&scaron;anje i hromatografska lipofilnost prirodnih stiril laktona 7-(+)-<br />goniofufurona, 7-epi-(+)-goniofufurona, krasalaktona B i C i dvadeset njihovih<br />novosintetizovanih derivata i analoga. U ranijim ispitivanjima pokazalo se da ova jedinjenja<br />imaju veliki biolo&scaron;ki potencijal jer pokazuju zapaženu citotoksičnost prema vi&scaron;e humanih<br />tumorskih ćelijskih linija. Hromatografsko pona&scaron;anje jedinjenja uglavnom je u skladu sa<br />njihovom strukturom. Ustanovljene su linearne veze između hromatografskih retencionih<br />konstanti i većine in silico parametara lipofilnosti. Primenom hemometrijske QSRR analize<br />utvrđeni su veoma dobri multi linearni regresioni prediktivni modeli kvantitativne zavisnosti<br />između eksperimentalno dobijene hromatografske retencione konstante, koja defini&scaron;e<br />retenciju jedinjenja u čistoj vodi i in silico molekulskih deskriptora odnosno strukture<br />jedinjenja. Lipofilnost jedinjenja ima najveći uticaj na njihove farmakokinetičke, tj. ADME<br />(apsorpcija, distribucija, metabolizam, eliminacija) osobine. Definisani su i statistički<br />potvrđeni najbolji multi linearni regresioni modeli zavisnosti farmakokinetičkih parametara<br />stiril laktona i od drugih molekulskih deskriptora. In vitro citotoksična aktivnost jedinjenja<br />evaluirana je prema četiri nove humane maligne ćelijske linije: kancer prostate (PC3), kancer debelog creva (HT-29), melanom (Hs294T), adenokancer pluća (A549). Najaktivnije<br />novosintetizovano jedinjenje je triciklični 4-fluorocinamatni analog, koji ispoljava<br />nanomolarnu aktivnost (IC₅₀ 2,1 nM) prema ćelijama melanoma i aktivniji je preko 2250 puta od komercijalnog antitumorskog agensa doksorubicina (DOX). SAR analizom utvrđena je zavisnost između strukture i biolo&scaron;ke aktivnosti jedinjenja. Molekulskim dokingom ispitana je veza stiril laktona i ciljanog proteina značajnog za kancer prostate. Jedinjenja sa visokom inhibitornom aktivno&scaron;ću prema ćelijama kancera prostate imaju visok doking skor i mogu graditi koordinativno-kovalentnu vezu sa Fe²+jonom prisutnim u aktivnom centru enzima. 3D-QSAR analizom, koja je izvedena metodama komparativnih polja CoMFA i CoMSIA, formiran je značajan prediktivni model između hemijske strukture i biolo&scaron;ke aktivnosti stiril laktona.</p> / <p>The behavior and the chromatographic lipophilicity natural styryl lactone 7-(+)-<br />goniofufurone, 7-epi-(+)-goniofufurone, crassalactones B and C and twenty of their newly<br />synthesized derivatives and analogs were examined using reverse-phase high performance liquid chromatography in the two solvent systems. In previous studies it has been shown that these compounds have great biological potential toward several human tumor cell lines. Chromatographic behavior of the compounds is generally in accordance with their structure. The relationships between the chromatographic retention constants and the majority of their in silico lipophilicity parameters are linear. The application of chemometric QSRR analysis determined very good multiple linear regression predictive models of quantitative correlation between experimentally obtained chromatographic retention constant, which determines the retention of the compound in pure water and in silico molecular descriptors, i.e. the structure of the compound. The lipophilicity of the compounds has a major influence on their pharmacokinetics, i.e. ADME (absorption, distribution, metabolism, elimination) properties. The best multi-linear regression models depending on the pharmacokinetic parameters of styryl lactone and other molecular descriptors have been defined and statistically validated. In vitro cytotoxic activity of the compounds was evaluated according to four novel human malignant cell lines: prostate cancer (PC3), colon cancer (HT-29), melanoma (Hs294T), lung adenocarcinom (A549). The most active compound was tricyclic 4-fluorocinnamic analog, which exhibits a nanomolar activity (IC50 2,1 nM) toward melanoma cells. This compound is over 2250 times more active than commercial antitumor agent doxorubicin (DOX). SAR analysis has revealed a correlation between the structure and the biological activity of the compounds. Using the molecular docking the relationship of the styryl lactone and the target protein important for prostate cancer was examined. The compounds with high inhibitory activity against prostate cancer cells have a high docking score and are capable to form a coordinative-covalent bond with a Fe2+ ion present in the active centre of the enzyme. 3DQSAR analysis, which was performed by methods of comparative CoMFA and CoMSIA fields, has formed a good predictive model between chemical structure and biological activity of the styryl lactone.</p>

Drug development is a long and expensive process. It is often not until potential drug candidates are administered to humans that accurate quantification of their pharmacokinetic characteristics is achieved. The goal of developing quantitative structure-pharmacokinetic relationships (QSPkRs) is to relate the molecular structure of a chemical entity with its pharmacokinetic characteristics. In this thesis artificial neural networks (ANNs) were used to construct in silico predictive QSPkRs for various pharmacokinetic parameters using different drug data sets. Drug pharmacokinetic data for all studies were taken from the literature. Information for model construction was extracted from drug molecular structure. Numerous theoretical descriptors were generated from drug structure ranging from simple constitutional and functional group counts to complex 3D quantum chemical numbers. Subsets of descriptors were selected which best modeled the target pharmacokinetic parameter(s). Using manual selective pruning, QSPkRs for physiological clearances, volumes of distribution, and fraction bound to plasma proteins were developed for a series of beta-adrenoceptor antagonists. All optimum ANN models had training and cross-validation correlations close to unity, while testing was performed with an independent set of compounds. In most cases the ANN models developed performed better than other published ANN models for the same drug data set. The ability of ANNs to develop QSPkRs with multiple target outputs was investigated for a series of cephalosporins. Multilayer perceptron ANN models were constructed for prediction of half life, volume of distribution, clearances (whole body and renal), fraction excreted in the urine, and fraction bound to plasma proteins. The optimum model was well able to differentiate compounds in a qualitative manner while quantitative predictions were mostly in agreement with observed literature values. The ability to make simultaneous predictions of important pharmacokinetic properties of a compound made this a valuable model. A radial-basis function ANN was employed to construct a quantitative structure-bioavailability relationship for a large, structurally diverse series of compounds. The optimum model contained descriptors encoding constitutional through to conformation dependent solubility characteristics. Prediction of bioavailability for the independent testing set were generally close to observed values. Furthermore, the optimum model provided a good qualitative tool for differentiating between drugs with either low or high experimental bioavailability. QSPkR models constructed with ANNs were compared with multilinear regression models. ANN models were shown to be more effective at selecting a suitable subset of descriptors to model a given pharmacokinetic parameter. They also gave more accurate predictions than multilinear regression equations. This thesis presents work which supports the use of ANNs in pharmacokinetic modeling. Successful QSPkRs were constructed using different combinations of theoretically-derived descriptors and model optimisation techniques. The results demonstrate that ANNs provide a valuable modeling tool that may be useful in drug discovery and development.

Drug development is a long and expensive process. It is often not until potential drug candidates are administered to humans that accurate quantification of their pharmacokinetic characteristics is achieved. The goal of developing quantitative structure-pharmacokinetic relationships (QSPkRs) is to relate the molecular structure of a chemical entity with its pharmacokinetic characteristics. In this thesis artificial neural networks (ANNs) were used to construct in silico predictive QSPkRs for various pharmacokinetic parameters using different drug data sets. Drug pharmacokinetic data for all studies were taken from the literature. Information for model construction was extracted from drug molecular structure. Numerous theoretical descriptors were generated from drug structure ranging from simple constitutional and functional group counts to complex 3D quantum chemical numbers. Subsets of descriptors were selected which best modeled the target pharmacokinetic parameter(s). Using manual selective pruning, QSPkRs for physiological clearances, volumes of distribution, and fraction bound to plasma proteins were developed for a series of beta-adrenoceptor antagonists. All optimum ANN models had training and cross-validation correlations close to unity, while testing was performed with an independent set of compounds. In most cases the ANN models developed performed better than other published ANN models for the same drug data set. The ability of ANNs to develop QSPkRs with multiple target outputs was investigated for a series of cephalosporins. Multilayer perceptron ANN models were constructed for prediction of half life, volume of distribution, clearances (whole body and renal), fraction excreted in the urine, and fraction bound to plasma proteins. The optimum model was well able to differentiate compounds in a qualitative manner while quantitative predictions were mostly in agreement with observed literature values. The ability to make simultaneous predictions of important pharmacokinetic properties of a compound made this a valuable model. A radial-basis function ANN was employed to construct a quantitative structure-bioavailability relationship for a large, structurally diverse series of compounds. The optimum model contained descriptors encoding constitutional through to conformation dependent solubility characteristics. Prediction of bioavailability for the independent testing set were generally close to observed values. Furthermore, the optimum model provided a good qualitative tool for differentiating between drugs with either low or high experimental bioavailability. QSPkR models constructed with ANNs were compared with multilinear regression models. ANN models were shown to be more effective at selecting a suitable subset of descriptors to model a given pharmacokinetic parameter. They also gave more accurate predictions than multilinear regression equations. This thesis presents work which supports the use of ANNs in pharmacokinetic modeling. Successful QSPkRs were constructed using different combinations of theoretically-derived descriptors and model optimisation techniques. The results demonstrate that ANNs provide a valuable modeling tool that may be useful in drug discovery and development.

This research developed validated QSPKR and PK-AS models for predicting human systemic PK properties of three, preselected, pharmacological classes of drugs, namely opioids, β-adrenergic receptor ligands (β-ARL) and β-lactam antibiotics (β-LAs) using pertinent human and animal systemic PK properties (fu,, CLtot, Vdss, fe) and their biologically relevant unbound counterparts from the published literature, followed by an assessment of the effect of different molecular descriptors on these PK properties and on the PK-AS slopes for CLtot and Vdss from two species (rat and dog). Lipophilicity (log (D)7.4) and molecular weight (MW) were found to be the most statistically significant and biologically plausible, molecular properties affecting the biologically relevant, systemic PK properties: For compounds with log (D)7.4 > -2.0 and MW < 350 D (e.g., most opioids and β-ARL), increased log (D)7.4 resulted in decreased fu and increased Vdssu, CLtotu and CLnonrenu, indicating the prevalence of hydrophobic interactions with biological membrane/proteins. As result, the final QSPKR models using log (D)7.4 provided acceptable predictions for fu, Vdssu, CLtotu and CLnonrenu. CLnonrenu and CLtotu. For both the datasets, inclusion of drugs undergoing extrahepatic clearance worsened the QSPKR predictions. For compounds with log (D)7.4 < -2.0 and MW > 350 D (e.g., β-LA), increased MW (leading to more hydrogen bond donors/acceptors) resulted in a decrease in fu, likely indicating hydrogen bonding interactions with plasma proteins. In general, it was more difficult to predict PK parameters for β-LAs, as their Vdssu approached plasma volume and CLrenu and CLnonrenu were low - as a result of their high hydrophilicity and large MW, requiring specific drug transporters for distribution and excretion. The PK-AS analysis showed that animal body size accounted for most of the observed variability (r2> 0.80) in systemic PK variables, with single species methods, particularly those using dog, gave the best predictions. The fu correction of PK variables improved goodness of fit and predictability of human PK. There were no apparent effects of molecular properties on the predictions. CLren, CLrenu, CLnonren, and CLnonrenu were the most difficult variables to predict, possibly due to the associated interspecies differences in the metabolism, renal and hepatobiliary drug transporters.